DarkDriving: A Real-World Day and Night Aligned Dataset for Autonomous Driving in the Dark Environment

2026-03-20 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Data Science & Analytics · Depth: Advanced, extended

Summary

DarkDriving is a new real-world benchmark dataset designed to advance research in low-light enhancement for autonomous driving. It comprises 9,538 precisely aligned day and night image pairs, collected from a 69-acre closed driving test field at Chang'an University. The dataset was created using a novel automatic Trajectory Tracking based Pose Matching (TTPM) method, which ensures alignment errors of only a few centimeters in both location and spatial content. Each image pair includes manually annotated 2D bounding boxes for objects, specifically cars. DarkDriving supports four perception tasks: low-light enhancement, generalized low-light enhancement, and low-light enhancement for 2D and 3D detection in dark environments. Experimental results demonstrate its effectiveness in improving image quality and promoting detection performance, even generalizing to other datasets like nuScenes.

Key takeaway

For research scientists developing vision-centric perception systems for autonomous vehicles, DarkDriving provides a critical resource for training and evaluating low-light enhancement models. You should leverage this dataset's precisely aligned day-night image pairs to develop more robust algorithms, particularly for improving 2D and 3D object detection in challenging nighttime conditions. This dataset can significantly reduce the day-night performance gap in your models.

Key insights

DarkDriving offers the first real-world dataset with precisely aligned day-night image pairs for autonomous driving.

Principles

• Precise day-night alignment is crucial for low-light enhancement.
• Real-world dynamic data improves generalization.
• Enhanced images boost perception system robustness.

Method

The Trajectory Tracking based Pose Matching (TTPM) method uses a high-precision map and autonomous vehicle control to ensure consistent day and night trajectories, followed by pose matching and human refinement for centimeter-level alignment.

In practice

• Use DarkDriving for supervised low-light enhancement training.
• Apply SNR-Aware for balanced image quality improvement.
• Fine-tune detectors with DarkDriving for robust 2D/3D detection.

Topics

• Autonomous Driving Datasets
• Low-light Image Enhancement
• Vision-centric Perception
• Object Detection
• Trajectory Tracking

Best for: Research Scientist, AI Researcher, AI Scientist, Computer Vision Engineer

Related on AIssential

• Making the Discrete Continuous: Synthetic RAW Augmentations for Fine-Grained Evaluation of Person Detection Performance in Low Light — Synthetic RAW augmentation can continuously evaluate low-light person detection, overcoming real data sparsity.
• DarkVesselNet: Multi-Modal Remote Sensing and Trajectory Reasoning for Dark Vessel Detection — Multi-modal remote sensing, combining SAR, optical, and AIS trajectory reasoning, significantly improves dark vessel detection capabilities.
• IG-Diff: Complex Night Scene Restoration with Illumination-Guided Diffusion Model — Illumination-guided diffusion models effectively restore complex night scenes by integrating illumination priors and synthetic data.
• Sensor2Sensor: Cross-Embodiment Sensor Conversion for Autonomous Driving — Sensor2Sensor converts monocular dashcam videos into multi-modal AV logs using a diffusion architecture and 4D Gaussian Splatting for data pairing.
• Learned Non-Maximum Suppression for 3D Object Detection — Learned filtering modules, D2D-Rescore and GossipNet3D, enhance 3D object detection by replacing heuristic NMS with detection-to-detection relation leveraging.
• Zero-Shot Semantic Re-Identification for Autonomous Driving: A VLM Baseline Study — Zero-shot semantic re-identification using VLMs generates textual descriptions for robust, interpretable identity matching in autonomous driving.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.